Abstract: Progress in artificial intelligence has made it easy to produce “Deep Fake” videos that are so realistic that even experts have trouble identifying them, and go on to spread virally, due to people’s susceptibility to content that appeals to their prejudices or fears, especially when forwarded by friends with whom they correspond regularly. It would seem that the hard sciences can do little to mitigate this problem, which has so much to do with psychology and human nature. But math and physics can be a significant part of the solution, by establishing in a hard-to-fake way a video’s time and place of origin, and that it has not been subsequently altered. An ordinary smartphone, if it is internet-connected, can be used to make rather hard-to-fake videos, and with the help of public randomness beacons, very hard-to-fake ones whose authenticity can be verified without needing to trust either the maker of the video or any centralized authority. A more serious problem is content that spreads virally despite containing no evidence at all of its provenance. Trusted open-source client-side scanning software and differential privacy techniques may offer a way to flag rapidly-spreading items for subsequent fact-checking without seriously compromising social media users’ privacy or freedom of speech.
Special Topics on Privacy And Public Auditability — Event 1
Starts: January 27, 2020Security and Privacy: cryptography, privacy